A Comparison of Numerical Techniques in Markov Modeling

Abstract

This paper presents several numerical methods which may be used to obtain the stationary probability vectors of Markovian models. An example of a nearly decomposable system is considered, and the results obtained by the different methods examined. A post mortem reveals why standard techniques often fail to yield the correct results. Finally, a means of estimating the error inherent in the decomposition of certain models is presented. © 1978, ACM. All rights reserved.